Deep Learning based CRES track and event reconstruction in Project 8

Project 8 is a next generation experiment aiming for a 40 meV neutrino mass sensitivity. This goal will be achieved by studying the tritium beta decay endpoint using the novel Cyclotron Radiation Emission Spectroscopy (CRES) technique. The endpoint can be resolved by magnetically trapping decay electrons and studying the spectrogram of their cyclotron radiation. These spectrograms are obtained by short-time Fourier transform of the time domain signals. In the spectrograms, signals of electron cyclotron radiation appear as tracks on top of background noise. We developed deep learning based models to find electron signals, and compared its performance to the pixel-grouping based baseline method for simulated data. This talk will discuss improvements in identifying electron tracks from noise background over the traditional baseline method. Efficiency gain and application in future phases of the experiment will also be highlighted.